If someone uses an async io in the filesystem backend, we are not granting credits on write/read responses, which slows down the clients. Jeremy, I'm not sure if this is blocker for 3.6.0, because both SMB2 and async io are not the defaults. If someone needs this, the patch can be applied on top of 3.6.0. However if Karolin is fine with picking it for 3.6.0, it would be nice to have. metze
The patch is currently in autobuild
Created attachment 6758 [details] Patch for v3-6-test
Comment on attachment 6758 [details] Patch for v3-6-test Great catch ! Thanks metze. This explains an unexpected slowdown with async io. Jeremy.
Re-assigning to Karolin for inclusion in 3.6.1. I agree, it is not a blocker for 3.6.0, but it should be one of the first patches applied to 3.6.1 :-). Cheers, Jeremy.
Pushed to v3-6-test. Will be included in 3.6.0. Closing out bug report. Thanks!
Reopened to add metze's regression fix ffbd1ed2797616226a7987efd4b4dd4275c8c9d0 from master. Jeremy.
Created attachment 7094 [details] Regression fix from metze for 3.6.2.
Comment on attachment 7094 [details] Regression fix from metze for 3.6.2. Looks good
Karolin, please pick for the next release
Pushed to v3-6-test. Closing out bug report. Thanks!